Why can't NVidia achieve this without spending tens of billions buying ARM?
Example: Nvidia would prefer certain specifications on an internal bus they'd use, but nobody else uses as a standard.
There doesn't appear to be sufficient profit margin in the industry for too many greenfield, non-reference custom implementations.
So if Nvidia-ARM sticks out a reference implementation or architecture choice that's beneficial to Nvidia, there's limited capability for others (Apple, Cavium/Marvell, Qualcomm, & Samsung?) to strike their own path, and they'll certainly be paying steeply to do so.
Having some advantage may be worth the purchase.